Augmented Skeleton Based Contrastive Action Learning with Momentum LSTM for Unsupervised Action Recognition

Rao, Haocong; Xu, Shihao; Hu, Xiping; Cheng, Jun; Hu, Bin

doi:10.1016/j.ins.2021.04.023

Cited by 146 publications

(125 citation statements)

References 9 publications

Supporting

Mentioning

101

Contrasting

Order By: Relevance

“…In this paper, we focus on the contrastive learning approach, which has recently shown promising results on several tasks [30,29,2,9,18,27,10,8,52]. In this method, we maximize agreement between representations of differently augmented views of the same data and contrast between representations coming from different images [2].…”

Section: Related Workmentioning

confidence: 99%

“…In [27] a momentumbased contrastive scheme was suggested, and [10] included a teacher-student distillation phase. Additional papers [26,52,60,49] introduced contrastive learning schemes for action classification of sequential inputs and non-sequential outputs. Motivated by these papers, we expand the contrastive learning framework to visual sequence-to-sequence predictions as in text recognition.…”

Section: Related Workmentioning

confidence: 99%

“…Contrastive learning techniques for self-supervised representation learning have recently demonstrated significant improvements on several semi-supervised computer vision applications, including image classification, object detection, and segmentation [9,27,18,8,26,52,10,60,49]. As illustrated in Fig.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Sequence-to-Sequence Contrastive Learning for Text Recognition

Aberdam

Litman²,

Tsiper³

et al. 2021

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Sequence-to-Sequence Contrastive Learning for Text Recognition

Aberdam

Litman²,

Tsiper³

et al. 2021

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

“…EMD is widely used in EEG and other physiological signal analysis [16] [17]. However, mode mixing is still a huge problem for EMD.…”

Section: A Emdmentioning

confidence: 99%

Analysis of Functional Brain Network in MDD Based on Improved Empirical Mode Decomposition With Resting State EEG Data

Shao

Sun

et al. 2021

IEEE Trans. Neural Syst. Rehabil. Eng.

Self Cite

View full text Add to dashboard Cite

At present, most brain functional studies are based on traditional frequency bands to explore the abnormal functional connections and topological organization of patients with depression. However, they ignore the characteristic relationship of electroencephalogram (EEG) signals in the time domain. Therefore, this paper proposes a network decomposition model based on Improved Empirical Mode Decomposition (EMD), it is suitable for timefrequency analysis of brain functional network. On the one hand, it solves the problem of mode mixing on original EMD method, especially on high-density EEG data. On the other hand, by building brain function networks on different intrinsic mode function (IMF), we can perform time-frequency analysis of brain function connections. It provides a new insight for brain function connectivity analysis of major depressive disorder (MDD). Experimental results found that the IMFs waveform decomposed by Improved EMD was more stable and the difference between IMFs was obvious, it indicated that the mode mixing can be effectively solved.Besides, the analysis of the brain network, we found that the changes in MDD functional connectivity on different IMFs, it may be related to the pathological changes for MDD. More statistical results on three network metrics proved that there were significant differences between MDD and normal controls (NC) group. In addition, the aberrant brain network structure of MDDs was also confirmed in the hubs characteristic. These findings may provide potential biomarkers for the clinical diagnosis of MDD patients.

show abstract

“…It has been proven that the long short-term memory (LSTM) network could solve the gradient vanishing problem of the basic RNN model [20], and the LSTM network performs better while processing long series problem such as language modeling [21]. Meanwhile, some modified model based on LSTM has been utilized in more scenarios such as real-time traffic flow prediction [22] and 3D action recognition [23,24]. Therefore, we use an LSTM network to obtain the TR-domain features of higher level by modeling the relationships between these flattened matrices.…”

Section: Introductionmentioning

confidence: 99%

Fatigue Detection with Spatial-Temporal Fusion Method on Covariance Manifolds of Electroencephalography

Zhao

Hou

et al. 2021

Entropy

Self Cite

View full text Add to dashboard Cite

With the increasing pressure of current life, fatigue caused by high-pressure work has deeply affected people and even threatened their lives. In particular, fatigue driving has become a leading cause of traffic accidents and deaths. This paper investigates electroencephalography (EEG)-based fatigue detection for driving by mining the latent information through the spatial-temporal changes in the relations between EEG channels. First, EEG data are partitioned into several segments to calculate the covariance matrices of each segment, and then we feed these matrices into a recurrent neural network to obtain high-level temporal information. Second, the covariance matrices of whole signals are leveraged to extract two kinds of spatial features, which will be fused with temporal characteristics to obtain comprehensive spatial-temporal information. Experiments on an open benchmark showed that our method achieved an excellent classification accuracy of 93.834% and performed better than several novel methods. These experimental results indicate that our method enables better reliability and feasibility in the detection of fatigued driving.

show abstract

Augmented Skeleton Based Contrastive Action Learning with Momentum LSTM for Unsupervised Action Recognition

Cited by 146 publications

References 9 publications

Sequence-to-Sequence Contrastive Learning for Text Recognition

Sequence-to-Sequence Contrastive Learning for Text Recognition

Analysis of Functional Brain Network in MDD Based on Improved Empirical Mode Decomposition With Resting State EEG Data

Fatigue Detection with Spatial-Temporal Fusion Method on Covariance Manifolds of Electroencephalography

Contact Info

Product

Resources

About